Research On Bearing Fault Diagnosis Method Based On Dynamic Bayesian Network

Rolling bearings are one of the most common and easily damaged parts in rotating machinery.Rolling bearing failure can cause abnormal vibration and noise of the machine,accelerate damage to the machine,and even cause serious accidents such as casualties.Monitoring and fault diagnosis of rolling bearings can quickly find and locate faults,formulate a reasonable and effective maintenance plan,and shorten the time of equipment failure,which is of great significance to the stable operation of the mechanical system.The operating environment of rolling bearings is complex and changeable,and fault signals are easily submerged by system disturbance and noise.At the same time,its fault mechanism is complicated and it is not easy to obtain a comprehensive and reliable fault causal logic relationship.The diagnosis method based on physical model is limited in practical application.However,the mechanical equipment generates a large amount of data containing the state of the rolling bearing during the operation,so the data-driven diagnosis method is widely used in the field of fault diagnosis.As a model that can express uncertain causal relationships,Bayesian networks have great advantages in solving the faults caused by the uncertainties and relationships of complex equipment.The Bayesian network can visualize multiple knowledge diagrams,and use probabilistic calculations for knowledge expression or reasoning.Its structure describes the causal relationship and conditional dependence between variables.The model has strong explanatory capabilities and its network structure is simple.The amount of training data is small and the calculation speed is faster.The main research contents of the thesis are as follows:(1)Carry out bearing failure analysis and feature index selection,extract highly sensitive feature indexes to characterize the operating state of the bearing,and input them into the network as observation nodes of the network;(2)Analyze the dependency relationship between the fault feature indicators according to the fault data,establish the Bayesian network topology,and learn the probability distribution of the network nodes,and establish a bearing fault diagnosis model based on the Bayesian network;on this basis,it is proposed The two-dimensional interpolation algorithm based on the JS divergence constraint interpolates the network conditional probability distribution table,and solves the problem of discontinuity of probability statistical factors in the probability distribution table due to incomplete failure data and the problem of more zero probability values,and improves the model’s performance Diagnostic accuracy rate;(3)Extend the Bayesian network in the time dimension,establish an online monitoring and diagnosis model based on the dynamic Bayesian network,and provide methods for network structure establishment and parameter learning.The model makes full use of the correlation between the time series of the monitoring data to describe the state transition during the bearing operation,and realizes the real-time monitoring and fault diagnosis of the bearing state through dynamic Bayesian network inference calculation.